Abstract: For the automatic detection of abnormal accidents on highways and the analysis of other video contents,the monitoring range offset detection is the precondition and basis.Due to the interference of moving objects,light and noises in the highway scenes,the existing detection methods show poor real-time performance and robustness.In order to solve these problems,an offset detection method based on corner set features is proposed.In this meth- od,first,false corners are removed according to the characteristics of the Taylor series,such as great extremism and few clusters of random noisy points.Then,the trained corner set features are used to accurately represent the image.On this basis,a matching is conducted by employing the corner set information rejecting the light interfer- ence,and both the cross correlation method and the dynamic threshold are adopted to realize the matching of the corner position and the number respectively,so as to avoid such interferences as light.Thus,the accurate detec- tion of deviation events is accomplished.The results of comparative experiments show that the proposed method im- proves the real-time performance and still maintains an average detection rate of 92%,which can better meet the requirement of highway monitoring.

Key words: corner set feature, offset detection, Taylor series, dynamic threshold, highway